Skip to content

fix(handle): notable lines on first spill, whole-line query - #578

Merged
justrach merged 1 commit into
release/v0.0.268from
cursor/handle-preview-554f
Aug 20, 2026
Merged

fix(handle): notable lines on first spill, whole-line query#578
justrach merged 1 commit into
release/v0.0.268from
cursor/handle-preview-554f

Conversation

@justrach

@justrach justrach commented Aug 20, 2026

Copy link
Copy Markdown
Owner

The #574 A/B on Grok 4.6 was worse on the scoreboard: forced cat of the 168 KB log took 4 API calls vs 3 on the previous binary. query=FAILED used an 80-byte pad and cut off request_id, so the model had to page again.

What changed

  • read_tool_result query returns the whole matching line.
  • The first handle payload includes a bounded excerpt of error-shaped lines (fail / error / panic / …), still under the 16 KiB threshold. The full blob stays on disk.
  • Live-ab usage regex accepts (cached, 0 cache writes).

Live A/B (Grok 4.6, n=2)

Before = 832236a (268 without this). After = this branch.

task before calls / in after calls / in success
f2b forced cat 3 / ~25k 2 / ~14.4k 2/2 both
f2 natural log 3 / ~24.5k 2 / ~14.3k 2/2 both
f1 bugfix 6 then 3 5 then 4 2/2 both (overlap)

After answers from the first spill (bash + completion). No pager turn.

Open in Web Open in Cursor 

query=FAILED used an 80-byte pad and cut off request_id, so Grok paid a
second pager turn (4 API calls vs 3 on the previous binary). Search now
returns the whole matching line, and the first handle payload includes a
bounded excerpt of error-shaped lines so a fat log's "what broke?" is
already in the spill.
@justrach
justrach marked this pull request as ready for review August 20, 2026 04:04
@justrach
justrach merged commit 06cf903 into release/v0.0.268 Aug 20, 2026
6 checks passed
@justrach
justrach deleted the cursor/handle-preview-554f branch August 20, 2026 04:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants